Most researchers assume that the cross race (CR) recognition deficit (e.g. the difficulty most people have in recognizing members of a race different from their own) is caused by reduced perceptual encoding expertise for CR faces. However, visual search and perceptual discrimination data I have collected suggest an alternative to this explanation. The CR recognition deficit may instead be caused by a feature selection process whereby race-specifying information is coded at the expense of other individuating information in CR faces. Experiments 1 and 2 replicate and extend finding from a perceptual discrimination task in which subjects who have difficulty recognizing CR faces are better able to discriminate among certain variants of CR faces than subjects who recognize CR faces accurately. Experiment 3 tests what differentiates the task used here, which reveals a CR discrimination advantage, from similar experiments revealing a CR disadvantage. More generally, these experiments seek to supplement domain-general models of object categories with a domain-specific understanding of the social concepts that drive specific feature selections in face categories.